EARTHQUAKE ENGINEERING & STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2017

Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/eqe.2887

Energy-compatible and spectrum-compatible (ECSC) ground

motion simulation using wavelet packets

Duruo Huang and Gang Wang*,†

Department of Civil and Environmental Engineering, Hong Kong University of Science and Technology, Clear Water
Bay, Kowloon, Hong Kong

SUMMARY
A stochastic ground-motion simulation and modification technique is developed to generate energy-
compatible and spectrum-compatible (ECSC) synthetic motions through wavelet packet characterization
and modification in both frequency and time domains. The ECSC method significantly advances traditional
spectral matching approaches, because it generates ground motions that not only match the target spectral
accelerations, but also match Arias intensity build-up and significant durations. The great similarity between
the ECSC simulated motions and the actual recorded motions is demonstrated through one-to-one compar-
ison of a variety of intensity measures. Extensive numerical simulations were also performed to validate the
performance of the ECSC ground motions through nonlinear analyses of elasto-plastic oscillators. The
ECSC method can be easily implemented in the generalized conditional intensity measure framework by
directly simulating a set of motions following a targeted distribution of multiple intensity measures. There-
fore, the ECSC method has great potential to be used in performance-based earthquake design and analysis.
Copyright © 2017 John Wiley & Sons, Ltd.

Received 23 March 2016; Revised 29 September 2016; Accepted 7 February 2017

KEY WORDS: synthetic ground motions; frequency- and time-domain modification; wavelet packet
analysis

1. INTRODUCTION

Performance-based earthquake design often requires evaluation of structural responses using

earthquake ground motion acceleration time histories. However, the number of available strong-
motion data is often insufficient. In regions of active seismicity, large-magnitude earthquakes at
design levels have very low probabilities of occurrence. Ground motion data are particularly sparse
in regions of low-to-moderate seismicity such as the Eastern United States, Australia and Hong Kong.
In order to analyze the seismic performance of structures, ground-motion time histories need to be
selected and modified from recorded strong-motion database [1, 2], or numerically generated using
either deterministic or stochastic approaches. In this paper, stochastic simulation refers to a class of
methods that directly generate ground-motion time histories using a few empirically calibrated param-
eters [3–11]. Comparing with physics-based deterministic numerical approaches [12, 13] and hybrid
methods [14], stochastic simulation does not consider the physics of fault rupture process and wave
propagation phenomena that require extensive computation. Yet, stochastic methods are particularly
cost effective for design purposes. It is possible to generate thousands of ground-motion time histories
within a few hours using stochastic simulation.

*Correspondence to: Gang Wang, Department of Civil and Environmental Engineering, Hong Kong University of
Science and Technology, Clear Water Bay, Kowloon, Hong Kong.
†
E-mail: gwang@ust.hk

Copyright © 2017 John Wiley & Sons, Ltd.

 D. HUANG AND G. WANG

Ground motion effects on structures are usually represented by spectral accelerations, which reflect
both amplitude and frequency content of a ground motion [15–18]. For design purposes, ground
motions can be stochastically simulated to be compatible with a prescribed response spectrum. A
notable example is SIMQKE, which iteratively adjusts the Fourier components of a modulated
Gaussian random process until the spectral amplitudes of the generated motion closely match a
target spectral shape [19]. One of the limitations of the above procedure is the frequency content of
the accelerograms cannot vary with time. The simulation is also not related to any specific
seismological environment.
However, the generation of the spectrum-compatible time histories is an ‘ill posed’ problem [20], in
which more than one ground-motion time history may be developed to match a target response
spectrum. Yet, these time histories may have totally different characteristics such as cumulative
energy, duration as well as nonstationary characteristics. These parameters have been found to be
important in the analysis of certain types of structures. For example, Chandramohan et al. [21, 22]
demonstrated that ground-motion duration is important in risk assessments of structural collapse;
Wang [23] these new stochastic modelsshowed that the Arias intensity (Ia), in addition to spectral
acceleration, is critical for estimating seismic slope displacements.
Therefore, rigorous ground-motion selection and modification process requires consideration of
multiple intensity measures (IMs) collectively. Following this line, a generalized conditional
intensity measure (GCIM) framework has been established for ground-motion selection [24–26].
The method is an extension of the conditional mean spectrum concept [27] and allows for selecting
ground motions conditional on multivariate IMs that are identified important in seismic response
analyses. The accuracy of the GCIM framework has been evaluated using nonlinear time history
analysis of simple structures [28, 29], which highlights the significance of the conditioning multiple
IMs in ground-motion selection process. However, none of the existing stochastic methods can
directly generate synthetic ground motions targeting at these multiple IMs.
In this study, a stochastic simulation and modification method is proposed to generate synthetic
ground motions whose response spectrum and Ia build-up process are compatible with specified
targets, resulting in so-called energy-compatible and spectrum-compatible (ECSC) ground motions.
The stochastic model adopted in this study was initially proposed by Yamamoto and Baker [30] to
simulate ground motions using wavelet packet transform (WPT). Compared with other wavelet-
based methods, WPT has basis functions that are orthogonal and localized in time and frequency
domains. The salient feature allows for ground-motion time histories to be flexibly adjusted in
frequency domain and time domain simultaneously. Therefore, their response spectrum and
cumulative energy can be modified. The authors will conduct systematic studies to validate the
simulated ECSC motions against real recorded motions using nonlinear oscillators. As the current
study can be regarded as stochastic modeling conditioned on given seismological environment,
response spectrum, duration and cumulative energy, the ECSC technique can be easily implemented
in the GCIM framework to directly simulate suitable ground motions for performance-based
earthquake design and analysis.

2. ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND-MOTION

SIMULATION AND MODIFICATION

2.1. Stochastic simulation of ground motions using wavelet packets

Artificial ground motions can be generated using many numerical methods. As one special form of
spectral presentation, the WPT was recently used to simulate nonstationary ground motions by
Yamamoto and Baker [30]. The method has also been recently extended for simulating spatially
correlated ground motions by Huang and Wang [31, 32]. There are several salient features of the
models: (i) Unlike the Gasparini–Vanmarcke model [19], these new stochastic models can
effectively characterize both amplitude and frequency nonstationarity of ground motions. The
nonstationarity of ground motions can greatly affect nonlinear structural responses [7, 11, 33]. (ii)

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

Model parameters have been empirically calibrated as functions of seismological variables such as
earthquake magnitude, source-to-site distance and site conditions. Therefore, synthetic ground
motions can be generated based on hazard-consistent scenarios and (iii) simulated motions from the
stochastic model have similar median and variability for a variety of IMs, which are consistent with
the IM-based ground-motion prediction models (GMPM) [34]. Therefore, the wavelet-packet based
stochastic models can be used for simulating future earthquakes.
The WPT is an advanced time–frequency analysis tool that decomposes a ground-motion time series
x(t) into a set of wavelet packets localized in the time (t) and frequency (f) domain. Specifically, for a
time series x(t), the wavelet packet coefficients cij;k are defined as:
∞
cij;k ¼ ∫ 0 xðt Þ ψ j;k
i
ðt Þdt (1)

where ψ ij;k ðt Þ is the wavelet packet basis function, which is chosen as the Meyer wavelet function
throughout this study, due to its good orthogonal property for time/frequency domain decomposition.
Given an acceleration time history in Figure 1(a), the time and frequency distribution of the squared


2

i

wavelet packet coefficients
cj;k
is the wavelet packet spectrum (WPS), as shown in Figure 1(b). In this
example, the time series contains 2N = 4096 (N = 12) data points evenly separated by a time interval
Δt = 0.01 s. The subscript j denotes the level of wavelet packet decomposition, which is chosen as
j = 8 throughout the study. The level of decomposition determines the frequency- and time-domain
resolution of the WPS. As shown in Figure 1(b), if the sampling frequency (Fs) of the time history
is 100 Hz; Nyquist frequency is Fs / 2 = 50 Hz. Therefore, the wavelet-packet transform divides the
frequency from 0 to the Nyquist frequency into bands of width Δfw = Fs / 2j + 1=0.1953 Hz. In other
words, the frequency axis is linearly discretized into 256 rows between 0 and 50 Hz. On the other
hand, the time interval between the centers of adjacent wavelet packets Δtw = 2 j  Δt = 2.56 s.

Figure 1. Wavelet packet spectrum, showing the distribution of squared wavelet packet coefficients of the
recorded acceleration time history at the San Valley—Roscoe Blvd station in the 1994 Northridge earth-
quake. [Colour figure can be viewed at wileyonlinelibrary.com]

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 D. HUANG AND G. WANG

Therefore, the WPS is divided into 2 N  j = 16 columns. Specifically, the notation cij;k denotes wavelet
packet coefficient localized in the i-th row and the k-th column of the WPS.
Using inverse transformation, time series x(t) can be precisely reconstructed via summation of
wavelet packets over all rows (i = 1 to 2 j) and columns (k = 1 to 2 N  j):
2 j 2 N j
xðt Þ ¼ ∑ ∑ cj;ki ψ j;ki ðt Þ: (2)
i¼1 k¼1

Additionally, the build-up of Ia with time is known as Husid function, defined as

π t
H ðt Þ ¼ ∫ jxðt Þj2 dt: (3)
2g ∞

The Ia is related to Husid function when the above integration is made for the entire time history. On
the other hand, it also equals to summation of all the squared coefficients in the WPS via Equation (4),

π ∞ π

2



Ia ¼ ∫0 jxðt Þj2 dt ¼ ∑ ∑
cij;k
: (4)
2g 2g i k

According to Yamamoto and Baker [30], 13 wavelet packet parameters have been developed to fully
characterize the statistical properties of the WPS, representing total energy, time- and frequency-domain
distribution and correlation of the wavelet packet coefficients as summarized in Table I. One may
consult Yamamoto and Baker [30], Huang and Wang [31, 32] for detailed information on the
definition of wavelet packet parameters. It should be noted that some of the wavelet packet
parameters are closely related to ground motion IMs. For example, the sum of squared wavelet
packet coefficients, termed as Eacc, represents the total wave energy. It is also directly related to the
Ia [35] by a constant factor 2g/π as shown in Equation (4). The functional form to predict these
wavelet packet parameters using seismological variables has been also provided in [30],

 
Y M w ; Rhyp ; Rrup ; V s30 ¼ α þ β1 M w þ β2 ln ðM w Þ þ β3 exp ðM w Þ
  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (5)
þ β4 Rhyp  Rrup þ β5 ln R2rup þ h2 þ β6 ln ðV s30 Þ þ η þ δ

where Y represents the natural logarithm of a wavelet packet parameter (except for the transformed
correlation coefficient ρ); Mw is the moment magnitude, Rrup is the rupture distance, Rhyp is the
hypocentral distance and Vs30 is the average shear-wave velocity in the top 30 m. The parameter h
denotes a fictitious depth in km. η and δ are intra- and inter-event residuals, respectively [30]. Given
an earthquake scenario, the prediction model can be used to generate ground-motion time histories.
Yamamoto [34] demonstrated that ground motions produced using the wavelet-packet simulation
have a variety of IMs consistent in both predicted median and variation to existing prediction model
in a wide range of magnitude, distance and site conditions. To model natural variability of ground
motions, residuals of the model prediction are also analyzed, including the inter-event standard

Table I. Summary of wavelet parameters and parameter groups.

Energy Time-domain mean and Frequency-domain mean and Time-frequency Randomness

parameters standard deviation standard deviation correlation parameter
Eacc E (t)major E (f)major ρ (t, f)major S (ξ)
E(a)major E (t)minor E (f)minor ρ (t, f)minor
S (t)major S (f)major
S (t)minor S (f)minor

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

deviation and the intra-event standard deviation [30, 34]. Spatial correlation model of these residuals
was also proposed for simulating spatially correlated ground motions [31, 32].

2.2. Energy-compatible and spectrum-compatible ground-motion simulation and modification

algorithm
Orthogonal property of the WPS allows for independent modification of time and frequency
characteristics of ground motions by adjusting wavelet packet coefficients. The wavelet packet
i
coefficients cj;k of the initial motion x(t) can be adjusted in the time and frequency domains
iteratively to match a target response spectrum and build-up of wave energy with time, respectively.
It is postulated that matching the build-up of Ia with time provides a critical constraint that can be
used to better regulate the ‘ill-posed’ spectrum-matching problem and results in more realistic
earthquake ground-motion time histories. The step-by-step algorithm of ECSC
simulation/modification method is illustrated as follows, as is also shown in Figure 2. In this
algorithm, ground motions are simulated and iteratively modified to match the target spectral
acceleration and Husid function.

2.2.1. Simulation of a seed motion. Knowing earthquake scenario parameters (Mw, R, Vs30), a target
response spectrum Sa target(f), a target Husid function H target(t) and a seed ground-motion time
history can be simulated using the following procedure. 
First, the median wavelet-packet parameters Y M w ; Rhyp ; Rrup ; V s30 are predicted using
Equation (5) given these earthquake scenario parameters. Knowing the target Ia value from the
target Husid function, several wavelet-packet parameters can be further improved. For example, the


2



wavelet-packet parameter Eacc is directly related to the target Ia via E acc ¼ ∑ ∑
cij;k
¼ 2g target
π Ia ,
i k
and the target value is assigned to the seed motion. An epsilon is used to measure the number of
standard deviation of the assigned Eacc value from the predicted median E acc via the following
equation:

Figure 2. Illustration of the energy-compatible and spectrum-compatible (ECSC) simulation/modification al-

gorithms. [Colour figure can be viewed at wileyonlinelibrary.com]

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 D. HUANG AND G. WANG

 
ε ¼ lnE acc  ln E acc =σ ln Eacc (6)

where σ ln Eacc (=0.85) is the standard deviation of Eacc in log scale. Because E(a)major is another energy
parameter closely correlated with Eacc, its value can be determined following the conditional mean
prediction:
ln E ðaÞmajor ¼ ln E ðaÞmajor þ ερ ln Eacc ; ln EðaÞmajor σ ln EðaÞmajor (7)

where ρ ln Eacc ; ln EðaÞmajor is the correlation coefficient between Eacc and E(a)major in log scale, which is
taken as 0.89 according to [34]. σ ln EðaÞmajor (=1.13) is the standard deviation of E(a)major in log scale.
Given the predicted and updated wavelet-packet parameters, the initial seed motion x(t) is then
generated using inverse wavelet-packet transform following Equation (2).
2.2.2. Adjustment on the wavelet packet spectrum. The seed motion x(t) is then modified to match the
target spectrum and target Husid function by multiplying the wavelet packet coefficients of the seed

motion using a weight matrix w(i, k). The modified motion can be written as y ðt Þ ¼
2 j 2 N j
∑ ∑ wði; k Þcj;k
i
ψ j;k
i
ðt Þ , where w(i, k) is the multiplier of wavelet packet coefficients cj;k
i
localized
i¼1 k¼1
around fi = iΔ fw, tk = kΔtw, i.e., the center of the i-th row and j-th column of the WPS. The
adjustment can be realized using a simple procedure to minimize the mismatch of spectrum and
energy build-up iteratively.
(a) Frequency-domain modification to match acceleration response spectrum
Using superscript (n) to denote quantities of the nth iteration, the simulated ground motion can be
2 j 2 N j
i ðnÞ
written as yðnÞ ðt Þ ¼ ∑ ∑ cj;k ψ j;k
i
ðt Þ. The weight of the WPS coefficients for the (n + 1)th iteration
i¼1 k¼1
can be determined according to the ratio of target spectral acceleration Sa target(fi) to simulated
spectral acceleration Sa(y(n)(t), fi) at oscillator frequency fi via Equation (8):

  Sa target ð f i Þ
wðnþ1Þ i; e
k ¼ for row i and column e
k: (8)
Sað y ðnÞ ðt Þ; f i Þ

Note the above modification is applied only to a specific WPS column number e k for each frequency
row i. Specifically, we first compute the instance te corresponding to the peak acceleration of a 5%
k
damped single-degree-of-freedom (SDOF) oscillator with natural frequency fi under ground motion
y(n)(t). This can be done during the process of calculating
 the response spectrum. Then, we find the
WPS column k in which t e is contained, i.e. k  1 Δt w < t e ≤ e
e e kΔ t w .
k k

 Because each row in the wavelet-packet spectrum covers a range of frequencies bounded by
ði1Þ F s iF s
2 jþ1
; 2 jþ1 , where i = 1 , 2 , ⋯ 2 j, Fs = 100, j = 8, modification of a wavelet packet in a row
could affecta range of frequencies. In this study, we match the spectral acceleration only at a single
frequency f i ¼ 2iFjþ1s in each row modification. Therefore, the modification process has fixed
frequency resolution, which is different relative to some of the competing spectral matching
algorithms [19, 37].
It is also worth pointing out that the frequency-domain modification changes response spectra
differently in different frequency ranges. A recent study [36] showed that scaling of response
spectral ordinates in the low to middle oscillator-frequency can be treated as being equivalent to
the scaling of the corresponding frequency in Fourier spectrum. However, the PGA and spectral
ordinates at high oscillator-frequencies (f ≥ 20 Hz) are controlled by the entire frequency band
instead of the high-frequency component of the ground motion. In this study, the PGA of the

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

simulated motion is adjusted directly to match the target PGA. The lowest and the highest frequency
for the row modification is set as 0.195 and 20 Hz, respectively (e.g. f1 = 0.195, f2 = 0.391, f3 = 0.586,
…,f102 = 19.922Hz). After the frequency-domain modification, the modified motion is written as
2 j 2 N j 2 j 2 N j
iðnÞ i ðnþ1Þ
y ðnþ1Þ ðt Þ ¼ ∑ ∑ w ðnþ1Þ cj;k ψ j;k
i
ðt Þ ¼ ∑ ∑ cj;k ψ j;k
i
ðt Þ.
i¼1 k¼1 i¼1 k¼1

(b) Time-domain modification to match Ia build-up

To match the Ia build-up process, the increment of Husid function of the target and simulated
motions should match within each WPS time interval. Accordingly, a piece-wise weighting function
can be determined to modulate the amplitude of the acceleration time history in each time interval
via Equation (9):

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðnþ2Þ ΔH target ðt k Þ
w ðt Þ ¼ for all column k ¼ 1; 2; ⋯2 N j and t∈½t k1 ; t k  (9)
ΔH ðnþ1Þ ðt k Þ

where ΔH (tk) = H(tk)  H(tk  1) is the increment of the Husid function over the WPS time
interval [tk  1,tk], and tk = kΔtw. Of course, the weighting function can also be smoothed in
practical use. After the time-domain modification, The modified motion then can be written as
y (n + 2)(t) = w (n + 2)(t) × y (n + 1)(t).
2.2.3. Evaluating compatibility of H(t) and Sa(f). To measure the compatibility of spectral
acceleration and Ia build-up, the following mean squared errors (MSEs) are defined to quantify the
relative difference in Sa and H(t) between the simulation and the target. Because H(t) is an
accumulative function, it is more sensible to compare the increments instead of the absolute values.

1 n 2
MSE Sa ¼ ∑ ð ln Saðyðt Þ; f i Þ  ln Sa target ðf i ÞÞ (10)
n i¼1
 2
1 m ΔH target ðt k Þ
MSE H ðtÞ ¼ ∑ 1 (11)
m k¼1 ΔH ðyðt Þ; t k Þ

Steps 2.2.2(a) and 2.2.2(b) will be iterated until the MSEs in Sa and H(t) reach a desired level. In the
end, baseline correction on the modified accelerogram will be performed to eliminate permanent drifts
in the ground velocity and displacement time histories.
It is worth mentioning that the proposed approach is different from many existing spectral matching
approaches that adjust ground-motion time histories either in the frequency domain [38] or in the time
domain [37, 39, 40]. Although the time-domain approach can result in a modified motion that preserves
the normalized Ia build-up process of the seed motion, it has been reported that the procedure cannot
control the value of Ia itself [41]. The method is only suitable for slight modification of the spectral
shape and relies heavily on the sensible choice of seed motions.

2.3. An illustrative example

In this section, an illustrative example of the ECSC method is presented using the ground motion
recorded at the El Centro Array #10 station in the 1979 Imperial Valley earthquake as a target motion.
Figure 3(a) shows the target acceleration time history, its WPS, 5%-damped acceleration response
spectrum and Ia build-up. Among them only response spectrum and Ia build-up are used as targets
in the ECSC modification process, while its acceleration time history and WPS are only illustrated
for reference. As discussed in section 2.2.1, a seed motion x(t) is first stochastically simulated using
seismological parameters of the target motion (Mw = 6.53, Rrup = 6.17 km, Rhyp = 28.13 km,
Vs30 = 203 m/s), as shown in Figure 3(b). It can be observed that neither spectral accelerations nor
H(t) can match those of the target motion, except that Ia of the seed motion matches the target value
as it is directly assigned. Mean squared errors of Sa and H(t) between the seed motion and the target

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 D. HUANG AND G. WANG

Figure 3. ECSC simulated ground motions, wavelet packet spectrum, response spectra and the build-up of
Arias intensity compared with the recorded motion at the El Centro Array #10 station in the 1979 Imperial
Valley earthquake.

are plotted in Figure 4, denoted as the 0th step. Subsequently, Figure 3(c) shows the simulated motion
after the 1st step frequency-domain modification, which reduces MSEs of Sa and H(t) as shown in
Figure 4. Then, the 2nd step, time-domain modification, is performed to match Ia build-up following
the procedure described section 2.2.2(b). Again, the 3rd step is frequency-domain modification and
the 4th step is time-domain modification. Figure 3(d) shows after the step 4, Husid function of the
simulated motion closely matches the target, although the spectrum compatibility needs further
improvement. The convergence profile in Figure 4 clearly demonstrated that the MSEs in Sa and H
(t) will not reduce further after 20 iterations. The ‘converged’ motion is termed as the ECSC
simulated ground motion, as shown in Figure 3(e).
In general, convergence of MSEs of Sa and H(t) is not strictly monotonic. Many of our tests showed
that MSEs become relatively stable after 20 iterations. It is desirable to get converged MSEs below
0.02. Otherwise, it is advised to use a different seed to perform the ECSC simulation/modification
again.

Figure 4. Convergence of Sa and H(t) in iterations.

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

By visual inspection, the ECSC simulated ground motion closely matches the response spectrum
and Ia build-up of the recorded motion. Apparently, they also have very similar peak ground
accelerations (PGAs) (around 0.23 g) and durations. Furthermore, change in time-domain and
frequency-domain characteristics of ground motions can be better represented in modification of
WPS in Figure 3. The wavelet packet coefficients of the seed motion appear to be larger and
densely populated at around 0.8 Hz, which may explain its higher spectral accelerations in that
frequency range. Frequency- and time-domain modification changed the amplitude and distribution
of wavelet packet coefficients. After iterations, WPS of the ECSC simulated motion shares a high
degree of overall similarity with that of the target motion in terms of amplitude and distribution.
It should be also noted that the generation of the initial seed motion x(t) in Step 0 is still a random
process, as described by Huang and Wang [31, 32]. Random realization of wavelet-packet parameters
will result in different seed motions using the same seismological variables. Figure 5 shows five
representative ECSC simulated ground motions, labeled as ‘Sim 1’ to ‘Sim 5’, generated from
different seed motions. By visual inspection, waveforms of the five simulations are similar to that of
the actual recorded motion in terms of acceleration, velocity and displacement time histories, even
though these time histories are not used as target in the simulation/modification process. It is
interesting to notice that all simulations generate a velocity pulse with a peak value greater than
40 cm/s at around 8 s, similar to that of the recorded motion. Moreover, the peak ground
displacements (PGDs) of three simulations are above 20 cm, similar to that of the recorded motion.

Figure 5. Five realizations of ECSC simulated ground motions, their response spectra and build-up of Ia
compared with ground motion recorded at the El Centro Array #10 station in the 1979 Imperial Valley
earthquake.

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 D. HUANG AND G. WANG

Figure 5 also shows comparison of response spectra, build-up of Ia and the Fourier spectra. In
general, all five simulated motions captured the frequency content and energy content of the
recorded motion reasonably well.

3. ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE SIMULATED GROUND

MOTION DATASET

The proposed ECSC simulation technique provides a viable approach to generate synthetic ground
motions that match both response spectra and energy build-up process with actual recorded motions.
The method can be used to ‘reproduce’ recorded ground motions to form a large dataset.
Performance of the ECSC method can be quantified by one-to-one comparison of several important
ground motion IMs between the recorded and simulated ground-motion datasets. Specifically, six
IMs are examined in this section including the PGA, peak ground velocity (PGV), PGD, significant
duration (D5–95), Ia and cumulative absolute velocity (CAV [42]).

3.1. Strong-motion dataset

A total of 562 ground-motion records from three large earthquakes in California are selected in the
validation test. They are the 1979 Imperial Valley earthquake, the 1989 Loma Prieta earthquake and
the 1994 Northridge earthquake. Table II provides information of these three events. Ground
motions are recorded in both strike-normal and strike-parallel directions (281 pairs) and are retrieved
from the Pacific Earthquake Engineering Research Center (PEER) Next Generation Attenuation
(NGA) strong-motion database. Figure 6 shows locations of seismograph stations and epicenters of
the three events. Using the ECSC method, each actual recorded motion is stochastically simulated
using seismological variables (M, R, Vs30) associated with that record. In this way, all 562
waveforms are ‘reproduced’ and referred to as the ECSC simulated dataset.
Figure 7 presents six representative recordings (in black line) from the NGA database and their
counterparts (in blue line) from the ECSC dataset. The target ground motions include records at (a)
the H-E13 and (b) the H-E04 station during the Imperial Valley earthquake, ground motions
recorded at (c) the Bear Valley and (d) the Foster City station during the Loma Prieta earthquake,
and ground motions recorded at (e) the Griffith Park and (f) the Sylmar station during the
Northridge earthquake. The rupture distance of these stations ranges from 5 to 52 km, while the
recorded PGA ranges from 0.12 to 0.68 g. Side-by-side comparisons of the actual recorded and

Table II. Earthquake events used to generate energy-compatible and spectrum-compatible (ECSC) dataset

Date Moment Number of

Event (yyyy/mm/dd) Magnitude Location recordings
Imperial Valley 1979/10/15 6.53 California 49
Loma Prieta 1989/10/18 6.93 California 82
Northridge 1994/01/17 6.69 California 150

simulated acceleration time histories, velocity time histories and displacement time histories are
provided. By visual inspection, the simulated and recorded acceleration time histories are found to
be similar in terms of both time and frequency characteristics. Response spectra and Ia build-up for
each pair of recorded and simulated ground motions are also compared in Figure 7. Both of them
match very well.

3.2. Comparison of ground-motion intensity measures

Ground motions in the ECSC dataset and their recorded counterparts in the NGA dataset are compared
in terms of seven important ground-motion IMs, including PGA, PGV, PGD, Ia, CAV and D5–95. The

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

Figure 6. Epicenters (stars) and locations of earthquake recording stations (triangles) used in this study.

residual term r lnIM is defined to quantify the relative difference between IMs (in the natural log scale)
using Equation (12):

r ln IM ¼ ln ðIM NGA Þ  ln ðIM ECSC Þ: (12)

Mean and standard deviation of r1n IM are summarized in Figure 8(a). On average, the mean of r ln IM
is in the range of 0 to 0.05 for PGA, PGV, D5–95, Ia and CAV, while the standard deviation is in the
range of 0 to 0.1. The results show that the ECSC simulation is overall unbiased to reproduce these
IMs. Majority of simulated IMs only deviate from the recorded motion by much less than 10%. On
the other hand, the mean and standard deviation of PGD residuals are 0.21 and 0.46, respectively. It
indicates that, on average, the PGDs of the ECSC ground motions are smaller than the recorded
counterparts by 23%, and variation is much larger than other IMs. On the other hand, the relative
difference between spectral accelerations can be quantified using:
   
r ln Sa ¼ ln SaðT ÞNGA  ln SaðT ÞECSC : (13)

Figure 8(b) shows the mean and standard deviation of rln Sa(T) for T = 0.01 to 5 s. It is observed that
the mean of residuals is in the range of 0.15 to 0.15. The standard deviation of residuals increases
with period, from less than 0.2 in the short periods (i.e. 0–0.5 s) to around 0.5 in the long period
(5 s). The results indicate the ECSC method cannot obtain the same accuracy in simulating the long
period characteristics of ground motions as it does for the short period.

4. TIME HISTORY ANALYSES OF ELASTO-PLASTIC OSCILLATORS USING ECSC AND

NGA DATASETS

4.1. Single-degree-of-freedom (SDOF) elasto-plastic oscillator

Nonlinear hysteretic analyses are often used to estimate lateral displacement and base shear demand of
structures under earthquake loading. In this section, we conduct one-to-one comparison of the dynamic
response of SDOF elasto-plastic oscillators using the simulated and the recorded ground motions.
Figure 9(a) shows the force–displacement relationship of a bilinear elasto-plastic oscillator with
post-yield stiffness equals 5% of the elastic stiffness. As shown in Figure 9(b), the oscillator is
designed according to the design response spectrum of site class C (Ss = 1.5 g, S1 = 0.6 g, Fa = 1.0,
Fv = 1.3) recommended by National Earthquake Hazards Reduction Program (NEHRP) [43]. If the
oscillator has an elastic fundamental period T0, its yield displacement Dy can be computed as:

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 D. HUANG AND G. WANG

Figure 7. Comparison of time histories, response spectra and Arias intensity build-up for six selected stations
from the NGA recorded dataset and the ECSC simulated dataset.

Dy ¼ SaðT 0 ÞðT 0 =2π Þ2 =R (14)

where R is the strength reduction factor that shows the level of ductility demands of the oscillator
model. Note that the term Sa(T0)(T0/2π)2 in Equation (12) equals to the spectral displacement,
which is defined as the maximum relative displacement of an elastic oscillator. For example, if
T0 = 1 s, R = 1, the yield displacement Dy is computed as 19.4 cm, which equals to the maximum
displacement of an elastic oscillator. In other words, the elasto-plastic oscillator remains in the linear

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

Figure 8. Comparison of NGA and ECSC dataset. (a) Box plot of residuals of PGA, PGV, PGD, Ia, CAV
and D5–95 and (b) mean and mean ± std of spectral acceleration residuals.

Figure 9. (a) Illustration of force–displacement relationship for a single-degree-of-freedom (SDOF) elasto-

plastic hysteretic system, and (b) NEHRP 5%-damped design response spectrum for Site Class C.

range. Increasing R reduces the yield displacement and drives the oscillator model into nonlinear range.
In this study, a total of 48 SDOF oscillators with a combination of eight strength reduction factor R
(i.e., 1, 2, 3 … 8) and six fundamental periods (i.e., 0.1, 0.2, 0.5, 1, 2, 5 s) are used.

4.2. Comparison of oscillator responses under the ECSC simulated and the recorded motions
Although the ECSC simulated motions resemble the recorded motions in terms of response spectrum
and accumulative energy, their capability to produce similar nonlinear oscillator responses should be
tested. Time-history analyses of the oscillators are conducted using a total of 562 recorded motions
and the same number of ECSC counterparts generated in Section 3.1. In time history analyses,
ground motion amplitudes are scaled to match the design spectrum at the oscillator-period T0
through a scaling factor f, which can be easily determined as f ¼ SaSa ðTð0TÞ 0 Þ . The scaled motions are
design

intensive to drive the oscillators into the nonlinear range when R > 1.
The maximum relative displacement and the absolute peak acceleration of the oscillators are
recorded for one-to-one comparison. Residuals rln Disp and rln Acc are defined to quantify the
difference in the maximum relative displacement and the absolute peak acceleration as follows,
r ln Disp ¼ ln ðDisp NGA Þ  ln ðDisp ECSC Þ (15)
r ln Acc ¼ ln ðAccNGA Þ  ln ðAccECSC Þ (16)

where subscript ‘NGA’ denotes that the oscillator response is calculated using a recorded motion from
the NGA database, while ‘ECSC’ denotes the response is from its ECSC simulated counterpart.

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 D. HUANG AND G. WANG

Figure 10 shows the mean and standard deviation of the displacement residuals versus strength
reduction factors. On average, the mean of displacement residuals is well bounded within ±0.1. The
standard deviation of residuals increases sharply from 0.05 to around 0.35, when R increases from 1
to 8. On the other hand, the mean and standard deviation of acceleration residuals are less than that
of the displacement residuals. Figure 11 shows that the mean of acceleration residuals is well
bounded within ±0.05, while the standard deviation gradually increases from 0.02 to around 0.2,
when R increases from 1 to 8. The above numerical analyses clearly demonstrate that the ECSC
ground motions can result in overall unbiased prediction of seismic responses of nonlinear
oscillators. The performance of the ECSC ground motions would not deteriorate significantly even
when highly nonlinear systems are involved.

5. APPLICATION OF ECSC SIMULATION IN GCIM FRAMEWORK

5.1. The generalized conditional intensity measure (GCIM) framework

As demonstrated in the previous sections, the ECSC simulation provides a viable approach to simulate
ground-motion time histories that match multiple target IMs. Obviously, matching the Ia build-up
process, a.k.a. the Husid function H(t), also implies the matching of several ground-motion duration

Figure 10. (a) Mean and (b) standard deviation of residuals for the maximum response displacements
calculated using scaled ground-motion datasets versus different strength reduction factors (R) and oscillator
periods (T0).

Figure 11. (a) Mean and (b) standard deviation of residuals for the maximum response acceleration
calculated using scaled ground-motion datasets versus different strength reduction factors (R) and oscillator
periods (T0).

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

measures, such as D5–75 and D5–95. Therefore, the ECSC method can be easily implemented in the
GCIM framework to generate suitable ground motions for performance-based earthquake design and
analysis. Different from the previous GCIM studies, the ground motions will be directly simulated
instead of being selected from a database.
Given an earthquake scenario, statistical distribution of the spectral accelerations, significant
durations (D5–75, D5–95) and Ia can be determined following the GCIM procedure [24–26]. Key to
this procedure, empirical cross-correlations between these IMs [e.g. [44–47]] need to be preserved
through multivariate random realizations. Figure 12(a) shows 60 randomly realized spectral
accelerations for an earthquake scenario (Mw = 7, Rrup = 10 km, strike-slip faulting) conditioned on
T = 1 s with epsilon = 0, following the GCIM procedure. Correspondingly, the cumulative
distribution functions (CDFs) of significant duration parameters, D5–75, D5–95 and Ia are shown in
Figure 12(b) for this scenario. In Figure 12, we also highlight a single realization of the spectral
acceleration and its corresponding values of significant durations (D5–75, D5–95) and Ia, which are
determined as 6.7 s, 12.5 s and 1.2 m/s, respectively. These values will be used to construct the
target Husid function for the ECSC simulation in the next section.

5.2. The target Husid function and ECSC simulation

According to [48], the normalized Husid function, as shown in Figure 13, can be well approximated by
a lognormal CDF via the following function:

Figure 12. (a) Distribution of spectral accelerations following the generalized conditional intensity measure
(GCIM) framework and a selected target Sa. (b) Cumulative distribution functions (CDFs) of D5–75, D5–95
and Arias intensity. The open circles represent corresponding values of the selected targets.

Figure 13. Normalized Husid function and fitted lognormal CDF.

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 D. HUANG AND G. WANG


ln ðt Þ  μ
H ðt Þ ¼ Φ I a (17)
σ

where Φ is the CDF of the standard normal distribution, μ and σ are two parameters to be determined,
which can be solved analytically if D5–75 and D5–95 are given. Define t5, t75 and t95 as the time
instances when 5, 75 and 95% of normalized Husid function are accumulated, which can be written as:
8  1

< t 5 ¼ exp ðμÞ exp σΦ ð0:05Þ 
>
t 75 ¼ exp ðμÞ exp σΦ1 ð0:75Þ : (18)
>
:  
1
t 95 ¼ exp ðμÞ exp σΦ ð0:95Þ

Therefore, D5–75 and D5–95 can be computed as follows:

(
   
D5-75 ¼ t 75  t 5 ¼ exp ðμÞ exp σΦ1 ð0:75Þ  exp σΦ1 ð0:05Þ

    : (19)
D5-95 ¼ t 95  t 5 ¼ exp ðμÞ exp σΦ1 ð0:95Þ  exp σΦ1 ð0:05Þ

Given the ratio of D5–75 and D5–95, σ in H(t) can be solved numerically via the following expression:
   
D595 exp σΦ1 ð0:95Þ  exp σΦ1 ð0:05Þ
¼    : (20)
D575 exp σΦ1 ð0:75Þ  exp σΦ1 ð0:05Þ

And μ in H(t) can be solved by Equation (21):

" #
D5-95
μ ¼ ln     : (21)
exp σΦ1 ð0:95Þ  exp σΦ1 ð0:05Þ

The above procedure converts the target significant durations and Ia generated from GCIM to the
target Husid function H(t) for the subsequent ECSC simulation.
Figure 14 compares the Sa and H(t) of an ECSC simulated ground motion with the targets
highlighted in Figure 12. Clearly, both response spectrum and Ia build-up of the simulated motion

Figure 14. (a) Target and simulated spectral acceleration, Arias intensity build-up. (b) Acceleration, velocity
and displacement time histories of the simulated ground motion.

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

agree well with the targets. Specifically, D5–75, D5–95 and Ia of the simulated motion are reported to be
6.3 s, 11.8 s and 1.15 m/s respectively. Obviously, the simulation procedure can be repeated for other
selected targets of Sa, D5–75, D5–95 and Ia build-up such that a full set of simulated motions can be
generated to follow the targeted distribution of these multiple IMs for a specific earthquake scenario.
This example demonstrates the great potential for the ECSC method to be implemented in the
GCIM framework.

6. CONCLUSIONS

In this study, a new ground-motion simulation and modification procedure was presented that allows
the generation of ECSC ground motions through wavelet packet characterization and modification.
The WPT has basis functions that are orthogonal and localized in time and frequency domains. The
salient feature allows for ground-motion time histories to be flexibly adjusted in frequency domain
and time domain simultaneously, thus, modifying their response spectrum and Ia build-up. The
procedure is based on three key steps, starting from prediction of a seed motion using seismological
constraints (Mw, R, Vs30), followed by iterative adjustments of the wavelet packet spectrum in both
time and frequency domain and, finally, evaluating compatibility of response spectrum and Ia build-
up of the simulated motion with targets.
The ECSC procedure was demonstrated by one-to-one comparison of several important IMs (e.g.
PGA, PGV, PGD, Ia) between simulated ECSC ground motions and actual recorded NGA ground
motions. Extensive numerical simulations were also performed to compare the nonlinear responses
of elasto-plastic oscillators using the simulated ECSC ground motions and recorded motions. The
tests validated that the general performance of the simulated ECSC ground motions are comparable
with that of the recorded motions. More examples to validate the ECSC ground motions using other
structural types or geotechnical systems can be found in Huang [49].
As the ECSC procedure can be regarded as stochastic ground-motion modeling conditioned on a
given seismological environment, response spectrum, duration and Ia build-up, the ECSC technique
can be easily implemented in the GCIM framework to directly simulate suitable ground motions for
performance-based earthquake design and analysis. The simulation can be repeated for many
realizations such that a collection of the simulated motions will follow the targeted distribution of
spectral accelerations, duration and Ia for a specific earthquake scenario. Different from the previous
GCIM studies, the ground motions are directly simulated instead of being selected from a database.
The ECSC simulation software can be accessed from the authors’ website at http://ihome.ust.hk/
~gwang/ ECSC_simulation.html.

ACKNOWLEDGMENTS
The study was supported by the Hong Kong Research Grants Council (RGC) through the General Research
Fund grant no 16213615 and University Grants Council (UGC) Research Infrastructure grant no.
IRS16EG15. The authors would like to acknowledge Jack Baker for sharing the source code for ground-
motion simulation at http://www.stanford.edu/~bakerjw/gm_simulation.html.

REFERENCES
1. Wang G. A ground motion selection and modification method capturing response spectrum characteristics and
variability of scenario earthquakes. Soil Dynamics and Earthquake Engineering 2011, 31:611–625.
2. Wang G, Youngs R, Power M, Li Z. Design ground motion library (DGML): an interactive tool for selecting earth-
quake ground motions. Earthquake Spectra 2015, 31(2):617–635.
3. Vanmarcke EH, Heredia-Zavoni E, Fenton GA. Conditional simulation of spatially correlated earthquake ground
motion. Journal of Engineering Mechanics 1993, 119(11):2333–2352.
4. Conte JP, Peng BF. Fully nonstationary analytical earthquake ground-motion model. Journal of Engineering
Mechanics 1997, 123(1):15–24.
5. Atkinson G, Boore D. Stochastic point-source modeling of ground motions in the Cascadia region. Seismological
Research Letters 1997, 68:74–85.

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 D. HUANG AND G. WANG

6. Atkinson G, Silva W. Stochastic modeling of California ground motions. Bulletin of the Seismological Society of
America 2000, 90:255–274.
7. Spanos P, Giaralis A, Politis N. Time frequency representation of earthquake accelerograms and inelastic structural
response records using the adaptive chirplet decomposition and empirical mode decomposition. Soil Dynamics and
Earthquake Engineering 2007, 27(7):675–689.
8. Rezaeian S, Kiureghian AD. A stochastic ground motion model with separable temporal and spectral
nonstationarities. Earthquake Engineering & Structural Dynamics 2008, 37:1565–1584.
9. Rezaeian S, Kiureghian AD. Simulation of synthetic ground motions for specified earthquake and site characteristics.
Earthquake Engineering & Structural Dynamics 2010, 39:1155–1180.
10. Zentner I. A procedure for simulating synthetic accelerograms compatible with correlated and conditional probabilis-
tic response spectra. Soil Dynamics and Earthquake Engineering 2014, 63:226–233.
11. Li Y, Conte J, Barbato M. Influence of time-varying frequency content in earthquake ground motions on seismic
response of linear elastic systems. Earthquake Engineering & Structural Dynamics 2016; in press.
12. Pitarka A, Irikura K, Iwata T, Sekiguchi H. Three dimensional simulation of the near-fault ground motion for
the 1995 Hyogo-ken Nanbu (Kobe), Japan, Earthquake. Bulletin of the Seismological Society of America 1998,
88(2):428–440.
13. He CH, Wang JT, Zhang CH, Jin F. Simulation of broad band seismic ground motions at dam canyons by using a
deterministic numerical approach. Soil Dynamics and Earthquake Engineering 2015, 76:136–144.
14. Graves RW, Pitarka A. Broadband ground-motion simulation using a hybrid approach. Bulletin of the Seismological
Society of America 2010, 100(5A):2095–2123.
15. Shome N, Cornell CA, Bazzurro P, Carballo JE. Earthquakes, records, and nonlinear responses. Earthquake Spectra
1998, 14(3):469–500.
16. Baker JW, Cornell CA. Spectral shape, epsilon and record selection. Earthquake Engineering & Structural Dynamics
2006, 35(9):1077–1095.
17. Chopra AK. Dynamics of Structures (4th edn). Prentice Hall, 2011.
18. Jayaram N, Lin T, Baker JW. A computationally efficient ground-motion selection algorithm for matching a target
response spectrum mean and variance. Earthquake Spectra 2011, 27(3):797–815.
19. Gasparini DA, Vanmarcke EH. Simulated Earthquake Motions Compatible with Prescribed Response Spectra.
Department of Civil Engineering: M.I.T, 1976.
20. Ostadan F., Mamoon S., Arango I. Effect of input motion characteristics on seismic ground response, Eleventh World
Conference on Earthquake Engineering, Paper No. 1924., Acapulco, Mexico, June 23–28, 1996.
21. Chandramohan R, Baker JW, Deierlein GG. Impact of hazard-consistent ground motion duration in structural col-
lapse risk assessment. Earthquake Engineering & Structural Dynamics 2016, 45(8):1357–1379.
22. Chandramohan R, Baker JW, Deierlein GG. Quantifying the influence of ground motion duration on structural col-
lapse capacity using spectrally equivalent records. Earthquake Spectra 2016, 32(2):927–950.
23. Wang G. Efficiency of scalar and vector intensity measures for seismic slope displacements. Frontiers of Structural
and Civil Engineering 2012, 6(1):44–52.
24. Bradley BA. A generalized conditional intensity measure approach and holistic ground-motion selection. Earthquake
Engineering & Structural Dynamics 2010, 39(12):1321–1342.
25. Bradley BA. The seismic demand hazard and importance of the conditioning intensity measure. Earthquake Engi-
neering & Structural Dynamics 2012, 41:1417–1437.
26. Tarbali K, Bradley BA. Ground motion selection for scenario ruptures using the generalized conditional intensity
measure (GCIM) method. Earthquake Engineering & Structural Dynamics 2015, 44(1):601–1621.
27. Baker JW. Conditional mean spectrum: tool for ground motion selection. Journal of Structural Engineering 2011,
137(3):322–331.
28. Kwong NS, Chopra AK. Evaluation of the exact conditional spectrum and generalized conditional intensity measure
methods for ground motion selection. Earthquake Engineering & Structural Dynamics 2015; online.
29. Kwong NS, Chopra AK, McGuire RK. Evaluation of ground motion selection and modification procedures using
synthetic ground motions. Earthquake Engineering & Structural Dynamics 2015, 44(11):1841–1861.
30. Yamamoto Y, Baker JW. Stochastic model for earthquake ground motion using wavelet packets. Bulletin of the Seis-
mological Society of America 2013, 103(6):3044–3056.
31. Huang D, Wang G. Stochastic simulation of regionalized ground motions using spatial cross-correlation model and
cokriging analysis. Earthquake Engineering & Structural Dynamics 2015, 44:775–794.
32. Huang D, Wang G. Regional-specific spatial cross-correlation model for stochastic simulation of ground-motion time
histories. Bulletin of the Seismological Society of America 2015, 105(1):272–284.
33. Yeh CH, Wen YK. Modeling of nonstationary ground motion and analysis of inelastic structural response. Structural
Safety 1990, 8:281–298.
34. Yamamoto Y. Stochastic model for earthquake ground motion using wavelet packets. Ph.D. Thesis, Department of
Civil and Environmental Engineering, Stanford University, Stanford, California, 2011.
35. Arias A. Measure of Earthquake Intensity, in: Seismic Design for Nuclear Power Plants, (Hansen RJ Ed.).
Massachusetts Institute of Technology Press: Cambridge, Massachusetts, 1970. p. 438–483.
36. Bora SS, Scherbaum F, Kuehn N, Stafford P. On the relationship between Fourier and response spectra: implications
for the adjustment of empirical ground-motion prediction equations (GMPEs). Bulletin of the Seismological Society
of America 2016, 106(3):1235–1253.

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe
 ENERGY-COMPATIBLE AND SPECTRUM-COMPATIBLE GROUND MOTIONS

37. Abrahamson NA. Non-stationary spectral matching. Seismological Research Letters 1992, 63(1):30.
38. Silva WJ, Lee K. WES RASCAL code for synthesizing earthquake ground motions. State-of-the-Art for Assessing
Earthquake Hazards in the United States. Report 24, U.S. Army Engineers Waterways Experiment Station, Misc.
Paper S-73-1, 1987.
39. Hancock J, Watson-Lamprey J, Abrahamson NA, Bommer JJ, Markatis A, McCoy E, Mendis R. An improved
method of matching response spectra of recorded earthquake ground motion using wavelets. Journal of Earthquake
Engineering 2006, 10:67–89.
40. Al Atik L, Abrahamson NA. An improved method for nonstationary spectral matching. Earthquake Spectra 2010, 26
(3):601–617.
41. Carlson C, Zekkos D, Athanasopoulos-Zekkos A, Hubler J. Impact of modification on ground motion characteristics
and geotechnical seismic analyses for a California site. Journal of Earthquake Engineering 2014, 18(5):696–713.
42. Du W, Wang G. A simple ground-motion prediction model for cumulative absolute velocity and model validation.
Earthquake Engineering & Structural Dynamics 2013, 42(8):1189–1202.
43. FEMA-356. Prestandard and commentary for the seismic rehabilitation of buildings. Federal Emergency Manage-
ment Agency, Washington DC, USA, 2000.
44. Bradley BA. Correlation of significant duration with amplitude and cumulative intensity measures and its use in
ground motion selection. Journal of Earthquake Engineering 2011, 15(6):809–832.
45. Baker JW, Jayaram N. Correlation of spectral acceleration values from NGA ground motion models. Earthquake
Spectra 2008, 24(1):299–317.
46. Du W, Wang G. Prediction equations for ground-motion significant durations using the NGA-West2 database.
Bulletin of the Seismological Society of America 2017, 107(1):319–333.
47. Wang G, Du W. Empirical correlations between cumulative absolute velocity and spectral accelerations from NGA
ground motion database. Soil Dynamics and Earthquake Engineering 2012, 43:229–236.
48. Stafford PJ, Sgobba S, Marano GC. An energy-based envelope function for stochastic simulation of earthquake
accelerograms. Soil Dynamics and Earthquake Engineering 2009, 29:1123–1133.
49. Huang D. Stochastic simulation of spatially correlated earthquake time histories, and energy-compatible and
spectrum-compatible ground motion modification using wavelet packets. Ph.D. thesis, Hong Kong University of Sci-
ence and Technology, 2016.

Copyright © 2017 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. (2017)
DOI: 10.1002/eqe